Developing diabetes ketoacidosis prediction using ANFIS model

This paper proposes the adaptive Neuro-fuzzy Inference System (ANFIS) to construct a diabetic ketoacidosis prediction model. Diabetic ketoacidosis results in large amount of ketones can be detected in urine through distinctive odour of acetone. Hence, urine odour analysis is able to diagnose diabetic ketoacidosis facilitating dianogstic test for diabetic patient. The Electronic Nose (E-nose) system consists of four metal oxide gas sensors was used to extract urine odour. Common process of diabetic diagnosis require patient to provide fasting urine for more accurate detection. Our work has shown both fasting and non-fasting urine are able to produce good diabetic detection accuracy through urine odour analysis. A total of 40 human subjects from CITO Laboratory, Semarang Central Java, Indonesia, involving 20 diabetic patients and 20 healthy subjects were used to build the prediction model. The proposed model has achieved at least 63% average accuracy in discriminating diabetic patient from healthy subject. When data preprocessing to average the training data samples was implemented, the detection performance was increased to above 93% The findings have shown promising results of using both fasting and non-fasting data samples for diabetic prediction. This is essential towards the flexibility of diabetic dianogstic test where it does not require patient to fast, therefore can be tested at anytime anywhere.